Genotyping by Sequencing (GBS) Workflows in SVS

         March 29, 2016

Did you know you can analyze your Genotyping by Sequencing (GBS) data in SVS? Well you can! You can combine tools for both GWAS quality control and analysis with tools for NGS data analysis to either identify SNPs in your dataset or to identify differences between populations or sub-species.

If your species has a reference sequence or even if you have a pseudo-map for your data, you can import it into our SNP & Variation Suite software and use the numerous tools available for quality control, association tests and visualization. If you also have a gene annotation source available you can perform variant classification for your SNPs.

This blog post will walk you through an example analysis of some GBS data for Maize (Zea mays) obtained from TASSEL. If you want to follow along, you can install the SVS Viewer available here and then download the example project.

SVS supports the import of many file formats including both hapmap (*.hmp) and VCF (*.vcf). Importing imputed data is also as easy as selecting the imputed file format. In this case, the VCF file was provided, but this VCF file only had genotype data available. If your VCF file had read depth, allelic depths or quality scores, this information could be used to filter the genotypes down to high-quality genotypes. See sections 2 & 3 of the Introduction to NGS Tutorial.

When you open up the example project you should see a window that looks like Fig. 1 below.

Maize Genotyping by Sequencing example project

Fig. 1. SVS project navigator for the Maize GBS Example Project (Click on images to enlarge)

Now there is a lot going on in this image. We’ll break it down into the following steps:

  1. Sample and SNP Quality Control
  2. Variant Classification
  3. Mixed Linear Model (EMMAX) Analysis
  4. Visualization

Sample and SNP Quality Control

To calculate sample statistics, double click on the “mdp_genotype – Genotypes(G_T) – Sheet 1” spreadsheet and go to Genotype > Genotype Sample Statistics. If you are examining the project using the SVS viewer, click on the right arrow (see Fig. 2) to expand the node and open the sample statistics results spreadsheet. 5 samples had a call rate of less than 80% and so these samples were excluded from further analysis.

Click on the indicated arrow to expand the project tree and see hidden spreadsheets

Fig. 2. Click on the indicated arrow to expand the project tree and see hidden spreadsheets (Click on images to enlarge)

SNPs were filtered by going to Genotype > Genotype Filtering by Marker and removing markers with:

  • Call Rate < 0.85
  • Number of alleles > 2
  • Alternate Allele Frequency (MAF) < 0.05 to keep common variants

Out of 3096 SNPs, the marker filtering removed 548 that did not pass the above criteria.

Finally, in order to create the most informative kinship matrix and subset of data most useful for principal component analysis (PCA), we want to filter out markers in high LD. This is done by LD pruning, Genotype > Quality Assurance and Utilities > LD Pruning. The defaults for LD pruning were used and this further reduced the number of markers to 2130 high quality, LD pruned SNPs.

To compute the kinship or genomic relationship matrix, go to Genotype > Quality Assurance and Utilities > GBLUP Genomic Relationship Matrix. The kinship matrix can be visualized in a heat map (see Fig. 3) and it is obvious that there is cryptic relatedness between the samples or lines of Maize. In fact, we know according to the documentation from Tassel and Panzea.org that these are inbred lines.

Heat map of relationship between maize lines

Fig. 3. Heat map of relationship between maize lines (click on images to enlarge)

Principal components can be computed by going to Genotype > Genotype Principal Component Analysis

Variant Classification

As this is GBS data from a VCF file, we can classify the variants based on their alternate alleles compared to the reference genome and using the location in the gene or exon. Variant classification is accessed in SVS from the genotype spreadsheet by going to DNA-Seq > Variant Classification. Use the Genotypes for Analysis spreadsheet for variant classification. These are all of the SNPs that pass quality control but have not been LD pruned. If you are interested in looking at rare variants you can use the original dataset.

This particular dataset only has 49 loss of function or missense variants in coding regions, see Fig. 4.

coding and missense variants

Fig. 4. Coding and missense variants 

Mixed Linear Model (EMMAX) Analysis

This dataset has three different phenotypes, for this example, we used ear height (EarHT) as the dependent variable. Using only the measurements from location A, we ran EMMAX and linear regression by going to Genotype > Mixed Linear Model Analysis. As we already computed the kinship matrix on the filtered data, we can select that spreadsheet to use for the analysis. After setting the parameters, the dialog looked like Fig. 5. If you have additional covariates you would like to correct for or a chromosome that is Hemizygous for males, you can correct for that as well by setting the proper parameters.

Set the parameters for Mixed Linear Model Analysis

Fig. 5. Set the parameters for Mixed Linear Model Analysis 

Two spreadsheets of output are created, one with the results for the linear regression (fixed effects only) analysis and one with the Single-Locus Mixed Model or EMMAX results. See Fig. 6.

Linear regression and EMMAX results for Location A with EarHT as dependent

Fig. 6. Linear regression and EMMAX results for Location A with EarHT as dependent 

Visualization of Results

The output from both models can be visualized in genomic context through GenomeBrowse in SVS. There are many ways to generate a GenomeBrowse viewer in SVS, for this example, from the project navigator click on the GenomeBrowse icon. See Fig. 7.

Create a GenomeBrowse window directly from the project navigator

Fig. 7. Create a GenomeBrowse window directly from the project navigator 

From the GenomeBrowse window, click on the Add button in the upper left hand corner and select Project from the top left. This brings up the project tree, click on P-Values from Linear Regression (node 94) and select -log10(P-Value) from the list on the right. Do the same for P-Values from Single-Locus Mixed Model (node 97). See Fig. 8.

Select results to visualize in GenomeBrowse

Fig. 8. Select results to visualize in GenomeBrowse 

After renaming the titles of the plots and coloring by chromosome, the GenomeBrowse window could look like Fig. 9.

GenomeBrowse visualization of results with a clear signal in Chromosome 1

Fig. 9. GenomeBrowse visualization of results with a clear signal in Chromosome 1

Many other tools are available for GBS data analysis in SVS, for more examples, please check out both the GWAS and DNA-Seq tutorials available for SVS. If you have any specific questions or would like to find out how SVS can aid in your data analysis please contact us at support@goldenhelix.com.

About Darby Kammeraad

Darby Kammeraad is the Director of Field Application Services at Golden Helix, joining the team in April of 2017. Darby graduated in 2016 with a master’s degree in Plant Sciences from Montana State University, where he also received his bachelor’s degree in Plant Biotechnology. Darby works on customer support and training. When not in the office, Darby is learning how to play guitar, hunting, fishing, snowboarding, traveling or working on a new recipe in the kitchen.

Leave a Reply

Your email address will not be published. Required fields are marked *